项目来源

俄罗斯科学基金(RSF)

项目主持人

Myasnikov Evgeny

项目受资助机构

Samara National Research University

项目编号

25-21-00413

立项年度

2025

立项时间

未公开

项目级别

国家级

研究期限

未知 / 未知

受资助金额

未知

学科

MATHEMATICS,INFORMATICS,AND SYSTEM SCIENCES-Intellectual data analysis and image recognition

学科代码

01-01-202

基金类别

未公开

распознавание действий человека ; обработка видео ; распознавание образов ; снижение размерности ; классификация ; система захвата движения ; human action recognition ; video processing ; pattern recognition ; dimensionality reduction ; classification ; motion capture system

参与者

未公开

参与机构

未公开

项目标书摘要：nnotation:The modern level of development of information technology allows you to come close to solving the most complex problems of video data analysis.Of particular interest among such tasks is the problem of creating intelligent systems for analyzing human actions.An effective solution to this problem will make it possible to offer many use cases:detecting crimes and dangerous behavior in order to ensure public safety,signaling attempts to unauthorized entry or manipulation of technical means of access control and management systems,improving the quality of human-machine interaction,increasing the safety of using virtual reality systems,tracking patient behavior for a quick response to the occurrence of life-threatening cases,monitoring children,interacting with smart home systems and much more.The growing demand for the creation of the above applications is supported by the active development of technical means of motion registration(motion capture systems,scene depth sensors,video recording means),as well as computing and algorithmic means.Although all of the above creates the prerequisites for increasing the effectiveness of automatic methods for analyzing human movement,the characteristics of the quality of solving the most important problems of motion analysis are still at an insufficient level for the widespread introduction of such systems.In addition,video data,most often used in such systems,have significant redundancy from the viewpoint of human movement analysis tasks,which causes additional costs in storing,transferring,processing and analyzing such data.The proposed project is aimed at improving the efficiency of automatic analysis of human movement.Within the framework of the project,it is proposed to develop a new approach to the construction of feature descriptions of human actions,which will eliminate the redundancy of the initial data without losing significant information about the action,will make it possible to reduce both the volume of transmitted and stored data,and the time of their subsequent processing,will allow solving the problems of motion analysis with a high level of quality.A feature of the proposed approach will be taking into account the spatial and temporal characteristics of movement,invariance to the time and scene scale,taking into account the angle of registration of an object during movement,using a model of a moving object taking into account the hierarchical nature of its various parts.The effectiveness of the proposed approach will be assessed by the quality of solving the problem of the classification of human actions.It is proposed to use not only known open datasets as initial data during the project execution,but also to create our own dataset using a motion capture system and video filming.To ensure the highest speed indicators,the proposed solutions are planned to be implemented using modern multicore architectures and graphics accelerators.Successful implementation of the project will contribute to the solution of the above applied problems,the development of the"computer vision"direction.The developed methods will find application as elements of artificial intelligence,in the creation of robotic devices,in virtual reality technologies.Note that according to a number of features,the tasks solved in the project will contribute to the development of end-to-end digital technologies"components of robotics and sensorics","technologies of virtual and augmented realities"and"neurotechnologies and artificial intelligence"included in the framework of the program"Digital Economy of the Russian Federation".Expected results:Expected results of the project:1.A new approach to the construction of feature-based descriptions of human actions.2.Methods for constructing feature descriptions of human actions from data obtained using a motion capture system(positional tracking)and data obtained using video filming.3.New methods for solving the problem of human action recognition according to their compact descriptions.4.Data set for research,obtained using a motion capture system(positional tracking)and video filming.5.Parallel implementations of methods for constructing feature descriptions of human actions and solving the problem of recognizing human actions based on their feature-based descriptions.6.Results of experimental studies of the developed methods.7.Based on the research results,it is planned to publish five scientific papers,four of which will be published in publications indexed in the Web of Science Core Collection or Scopus databases.The scientific significance of the results lies in the development of the direction of"computer vision",the theory of reducing the dimension of multidimensional data.The applied significance of the results lies in increasing the efficiency of solving problems of processing and analyzing human movement.It should be noted that the expected results can serve as a basis for the further creation of intelligent applied systems based on the analysis of human actions in the field of safety,robotics,medicine,etc.

Application Abstract: Annotation:The modern level of development of information technology allows you to come close to solving the most complex problems of video data analysis.Of particular interest among such tasks is the problem of creating intelligent systems for analyzing human actions.An effective solution to this problem will make it possible to offer many use cases:detecting crimes and dangerous behavior in order to ensure public safety,signaling attempts to unauthorized entry or manipulation of technical means of access control and management systems,improving the quality of human-machine interaction,increasing the safety of using virtual reality systems,tracking patient behavior for a quick response to the occurrence of life-threatening cases,monitoring children,interacting with smart home systems and much more.The growing demand for the creation of the above applications is supported by the active development of technical means of motion registration(motion capture systems,scene depth sensors,video recording means),as well as computing and algorithmic means.Although all of the above creates the prerequisites for increasing the effectiveness of automatic methods for analyzing human movement,the characteristics of the quality of solving the most important problems of motion analysis are still at an insufficient level for the widespread introduction of such systems.In addition,video data,most often used in such systems,have significant redundancy from the viewpoint of human movement analysis tasks,which causes additional costs in storing,transferring,processing and analyzing such data.The proposed project is aimed at improving the efficiency of automatic analysis of human movement.Within the framework of the project,it is proposed to develop a new approach to the construction of feature descriptions of human actions,which will eliminate the redundancy of the initial data without losing significant information about the action,will make it possible to reduce both the volume of transmitted and stored data,and the time of their subsequent processing,will allow solving the problems of motion analysis with a high level of quality.A feature of the proposed approach will be taking into account the spatial and temporal characteristics of movement,invariance to the time and scene scale,taking into account the angle of registration of an object during movement,using a model of a moving object taking into account the hierarchical nature of its various parts.The effectiveness of the proposed approach will be assessed by the quality of solving the problem of the classification of human actions.It is proposed to use not only known open datasets as initial data during the project execution,but also to create our own dataset using a motion capture system and video filming.To ensure the highest speed indicators,the proposed solutions are planned to be implemented using modern multicore architectures and graphics accelerators.Successful implementation of the project will contribute to the solution of the above applied problems,the development of the"computer vision"direction.The developed methods will find application as elements of artificial intelligence,in the creation of robotic devices,in virtual reality technologies.Note that according to a number of features,the tasks solved in the project will contribute to the development of end-to-end digital technologies"components of robotics and sensorics","technologies of virtual and augmented realities"and"neurotechnologies and artificial intelligence"included in the framework of the program"Digital Economy of the Russian Federation".Expected results:Expected results of the project:1.A new approach to the construction of feature-based descriptions of human actions.2.Methods for constructing feature descriptions of human actions from data obtained using a motion capture system(positional tracking)and data obtained using video filming.3.New methods for solving the problem of human action recognition according to their compact descriptions.4.Data set for research,obtained using a motion capture system(positional tracking)and video filming.5.Parallel implementations of methods for constructing feature descriptions of human actions and solving the problem of recognizing human actions based on their feature-based descriptions.6.Results of experimental studies of the developed methods.7.Based on the research results,it is planned to publish five scientific papers,four of which will be published in publications indexed in the Web of Science Core Collection or Scopus databases.The scientific significance of the results lies in the development of the direction of"computer vision",the theory of reducing the dimension of multidimensional data.The applied significance of the results lies in increasing the efficiency of solving problems of processing and analyzing human movement.It should be noted that the expected results can serve as a basis for the further creation of intelligent applied systems based on the analysis of human actions in the field of safety,robotics,medicine,etc.